文档介绍：PATTERN IDENTIFICATION OF TIME
SERIES EVENTS: A PARALLEL
IC ALGORITHM APPROACH
Stellios Keskinidis
Bachelor of Information Technology, Honours
Faculty of Information Technology
Queensland University of Technolgy
November 2002
Abstract
ic Algorithms are an adaptive search technique based on the princi-
ples and mechanisms of natural selection. Their highly parallelizable nature
provides an increased amount of resources for solving problems. This disser-
tation discusses the implementation of a parallel ic algorithm for the
Time Series Data Mining Framework.
The Time Series Data Mining Framework is a novel method for dis-
covering patterns characteristic of time series events. A ic algorithm
searches an underlying geometric space for patterns by applying a clustering
technique. A statistical method measures the quality of each pattern.
Domain position of the geometric space opens the opportunity
for parallel execution. Running the ic algorithm search in parallel in-
creases the amount putational resources in solving the problem. It
yields a speedup and increases the probability of finding better solutions.
This dissertation also draws attention to the advantages of domain -
position for parallelizing ic algorithms where the underlying data is a
geometric representation.
Acknowledgement
I would like to thank the Faculty of Information Technology at the
Queensland University of Technology for giving me the opporunity -
plete Honours. I am thankful to Dr On Wong for taking the role as my
supervisor and Dr Maolin Tang for his thoughts on ic algorithms. I
would also like to thank Ben Fowler and Jimmy Louca for there insights and
suggestions.
3
Contents
1 Introduction 9
Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . 10
Investigation . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Dissertation Outline . . . . . . . . . . . . . . . . . . . . . . . 11
2 Overview of ic Algorithms 12